Aim Dynamic range models are proposed to investigate species distributions and to project range shifts under climate change. They are based upon the Hutchinsonian niche theory, specifying that the occurrence of a species in an environmental space should be limited to positions where the intrinsic growth rate is positive. Evaluating population growth rate is, however, difficult for physiologically structured populations, such as forest stands, owing to size-induced individual variation in performance. Therefore, we still have a limited understanding of which aspect of tree demography contributes the most to their geographical range limit. We develop an index of demographic performance for size-structured populations and study its variation across a climatic gradient. We then investigate the relationship between the demographic performance index and species distribution. Location North America (57-124 degrees W, 26-52 degrees N). Time period 1963-2010. Major taxa studied Fourteen tree species. Methods We represent forest dynamics with a size-structured population model and neighbourhood competition with the perfect plasticity approximation. We then derive the lifetime reproduction per individual, R0, in the absence of density dependence. Using forest inventory data, we assess how tree demography for each species varies with climate. We test the model by comparing R0 and the probability of occurrence within species ranges. Results We find that both growth and mortality rates vary across species distributions, but climate explains little of the observed variation. Individual size and neighbourhood competition are the primary explanatory variables of tree demography. Finally, we find that R0 relates weakly to the probability of occurrence, with no systematic decline in population growth rates towards the range limits. Main conclusions Spatial and size-induced variation in tree growth and mortality do not explain range limits and are insufficient to enable an understanding of tree dynamics. We propose that phenomena perceived mostly at the metapopulation scale should also be considered.